Sunghoon Im

Jinwoo Bae, Sungho Moon, Sunghoon Im

Self-supervised monocular depth estimation has been widely studied recently. Most of the work has focused on improving performance on benchmark datasets, such as KITTI, but has offered a few experiments on generalization performance. In this paper, we investigate the backbone networks (e.g. CNNs, Transformers, and CNN-Transformer hybrid models) toward the generalization of monocular depth estimation. We first evaluate state-of-the-art models on diverse public datasets, which have never been seen during the network training. Next, we investigate the effects of texture-biased and shape-biased representations using the various texture-shifted datasets that we generated. We observe that Transformers exhibit a strong shape bias and CNNs do a strong texture-bias. We also find that shape-biased models show better generalization performance for monocular depth estimation compared to texture-biased models. Based on these observations, we newly design a CNN-Transformer hybrid network with a multi-level adaptive feature fusion module, called MonoFormer. The design intuition behind MonoFormer is to increase shape bias by employing Transformers while compensating for the weak locality bias of Transformers by adaptively fusing multi-level representations. Extensive experiments show that the proposed method achieves state-of-the-art performance with various public datasets. Our method also shows the best generalization ability among the competitive methods.

Jaeyeul Kim, Jungwan Woo, Ukcheol Shin et al.

Understanding the motion states of the surrounding environment is critical for safe autonomous driving. These motion states can be accurately derived from scene flow, which captures the three-dimensional motion field of points. Existing LiDAR scene flow methods extract spatial features from each point cloud and then fuse them channel-wise, resulting in the implicit extraction of spatio-temporal features. Furthermore, they utilize 2D Bird's Eye View and process only two frames, missing crucial spatial information along the Z-axis and the broader temporal context, leading to suboptimal performance. To address these limitations, we propose Flow4D, which temporally fuses multiple point clouds after the 3D intra-voxel feature encoder, enabling more explicit extraction of spatio-temporal features through a 4D voxel network. However, while using 4D convolution improves performance, it significantly increases the computational load. For further efficiency, we introduce the Spatio-Temporal Decomposition Block (STDB), which combines 3D and 1D convolutions instead of using heavy 4D convolution. In addition, Flow4D further improves performance by using five frames to take advantage of richer temporal information. As a result, the proposed method achieves a 45.9% higher performance compared to the state-of-the-art while running in real-time, and won 1st place in the 2024 Argoverse 2 Scene Flow Challenge. The code is available at https://github.com/dgist-cvlab/Flow4D.

EungGu Kang, Byeonghun Lee, Sunghoon Im et al.

Multi frame super-resolution(MFSR) achieves higher performance than single image super-resolution (SISR), because MFSR leverages abundant information from multiple frames. Recent MFSR approaches adapt the deformable convolution network (DCN) to align the frames. However, the existing MFSR suffers from misalignments between the reference and source frames due to the limitations of DCN, such as small receptive fields and the predefined number of kernels. From these problems, existing MFSR approaches struggle to represent high-frequency information. To this end, we propose Deep Burst Multi-scale SR using Fourier Space with Optical Flow (BurstM). The proposed method estimates the optical flow offset for accurate alignment and predicts the continuous Fourier coefficient of each frame for representing high-frequency textures. In addition, we have enhanced the network flexibility by supporting various super-resolution (SR) scale factors with the unimodel. We demonstrate that our method has the highest performance and flexibility than the existing MFSR methods. Our source code is available at https://github.com/Egkang-Luis/burstm

Seunghun Lee, Wonhyeok Choi, Changjae Kim et al.

In this paper, we present a direct adaptation strategy (ADAS), which aims to directly adapt a single model to multiple target domains in a semantic segmentation task without pretrained domain-specific models. To do so, we design a multi-target domain transfer network (MTDT-Net) that aligns visual attributes across domains by transferring the domain distinctive features through a new target adaptive denormalization (TAD) module. Moreover, we propose a bi-directional adaptive region selection (BARS) that reduces the attribute ambiguity among the class labels by adaptively selecting the regions with consistent feature statistics. We show that our single MTDT-Net can synthesize visually pleasing domain transferred images with complex driving datasets, and BARS effectively filters out the unnecessary region of training images for each target domain. With the collaboration of MTDT-Net and BARS, our ADAS achieves state-of-the-art performance for multi-target domain adaptation (MTDA). To the best of our knowledge, our method is the first MTDA method that directly adapts to multiple domains in semantic segmentation.

Jinwoo Bae, Kyumin Hwang, Sunghoon Im

Monocular depth estimation has been widely studied, and significant improvements in performance have been recently reported. However, most previous works are evaluated on a few benchmark datasets, such as KITTI datasets, and none of the works provide an in-depth analysis of the generalization performance of monocular depth estimation. In this paper, we deeply investigate the various backbone networks (e.g.CNN and Transformer models) toward the generalization of monocular depth estimation. First, we evaluate state-of-the-art models on both in-distribution and out-of-distribution datasets, which have never been seen during network training. Then, we investigate the internal properties of the representations from the intermediate layers of CNN-/Transformer-based models using synthetic texture-shifted datasets. Through extensive experiments, we observe that the Transformers exhibit a strong shape-bias rather than CNNs, which have a strong texture-bias. We also discover that texture-biased models exhibit worse generalization performance for monocular depth estimation than shape-biased models. We demonstrate that similar aspects are observed in real-world driving datasets captured under diverse environments. Lastly, we conduct a dense ablation study with various backbone networks which are utilized in modern strategies. The experiments demonstrate that the intrinsic locality of the CNNs and the self-attention of the Transformers induce texture-bias and shape-bias, respectively.

Wonhyeok Choi, Sunghoon Im

In this paper, we present a new MTL framework that searches for structures optimized for multiple tasks with diverse graph topologies and shares features among tasks. We design a restricted DAG-based central network with read-in/read-out layers to build topologically diverse task-adaptive structures while limiting search space and time. We search for a single optimized network that serves as multiple task adaptive sub-networks using our three-stage training process. To make the network compact and discretized, we propose a flow-based reduction algorithm and a squeeze loss used in the training process. We evaluate our optimized network on various public MTL datasets and show ours achieves state-of-the-art performance. An extensive ablation study experimentally validates the effectiveness of the sub-module and schemes in our framework.

Hojin Kim, Seunghun Lee, Sunghoon Im

In this paper, we present offline-to-online knowledge distillation (OOKD) for video instance segmentation (VIS), which transfers a wealth of video knowledge from an offline model to an online model for consistent prediction. Unlike previous methods that having adopting either an online or offline model, our single online model takes advantage of both models by distilling offline knowledge. To transfer knowledge correctly, we propose query filtering and association (QFA), which filters irrelevant queries to exact instances. Our KD with QFA increases the robustness of feature matching by encoding object-centric features from a single frame supplemented by long-range global information. We also propose a simple data augmentation scheme for knowledge distillation in the VIS task that fairly transfers the knowledge of all classes into the online model. Extensive experiments show that our method significantly improves the performance in video instance segmentation, especially for challenging datasets including long, dynamic sequences. Our method also achieves state-of-the-art performance on YTVIS-21, YTVIS-22, and OVIS datasets, with mAP scores of 46.1%, 43.6%, and 31.1%, respectively.

Jaeyeul Kim, Jungwan Woo, Jeonghoon Kim et al.

In the realm of LiDAR-based perception, significant strides have been made, yet domain generalization remains a substantial challenge. The performance often deteriorates when models are applied to unfamiliar datasets with different LiDAR sensors or deployed in new environments, primarily due to variations in point cloud density distributions. To tackle this challenge, we propose a Density Discriminative Feature Embedding (DDFE) module, capitalizing on the observation that a single source LiDAR point cloud encompasses a spectrum of densities. The DDFE module is meticulously designed to extract density-specific features within a single source domain, facilitating the recognition of objects sharing similar density characteristics across different LiDAR sensors. In addition, we introduce a simple yet effective density augmentation technique aimed at expanding the spectrum of density in source data, thereby enhancing the capabilities of the DDFE. Our DDFE stands out as a versatile and lightweight domain generalization module. It can be seamlessly integrated into various 3D backbone networks, where it has demonstrated superior performance over current state-of-the-art domain generalization methods. Code is available at https://github.com/dgist-cvlab/MultiDensityDG.

SungHo Moon, JinWoo Bae, SungHoon Im

Research on monocular 3D object detection is being actively studied, and as a result, performance has been steadily improving. However, 3D object detection performance is significantly reduced when applied to a camera system different from the system used to capture the training datasets. For example, a 3D detector trained on datasets from a passenger car mostly fails to regress accurate 3D bounding boxes for a camera mounted on a bus. In this paper, we conduct extensive experiments to analyze the factors that cause performance degradation. We find that changing the camera pose, especially camera orientation, relative to the road plane caused performance degradation. In addition, we propose a generalized 3D object detection method that can be universally applied to various camera systems. We newly design a compensation module that corrects the estimated 3D bounding box location and heading direction. The proposed module can be applied to most of the recent 3D object detection networks. It increases AP3D score (KITTI moderate, IoU $> 70\%$) about 6-to-10-times above the baselines without additional training. Both quantitative and qualitative results show the effectiveness of the proposed method.

Woo Kyoung Han, Sunghoon Im, Jaedeok Kim et al. · nvidia

We propose a practical approach to JPEG image decoding, utilizing a local implicit neural representation with continuous cosine formulation. The JPEG algorithm significantly quantizes discrete cosine transform (DCT) spectra to achieve a high compression rate, inevitably resulting in quality degradation while encoding an image. We have designed a continuous cosine spectrum estimator to address the quality degradation issue that restores the distorted spectrum. By leveraging local DCT formulations, our network has the privilege to exploit dequantization and upsampling simultaneously. Our proposed model enables decoding compressed images directly across different quality factors using a single pre-trained model without relying on a conventional JPEG decoder. As a result, our proposed network achieves state-of-the-art performance in flexible color image JPEG artifact removal tasks. Our source code is available at https://github.com/WooKyoungHan/JDEC.

Seunghun Lee, Jiwan Seo, Kiljoon Han et al.

In this paper, we introduce the Context-Aware Video Instance Segmentation (CAVIS), a novel framework designed to enhance instance association by integrating contextual information adjacent to each object. To efficiently extract and leverage this information, we propose the Context-Aware Instance Tracker (CAIT), which merges contextual data surrounding the instances with the core instance features to improve tracking accuracy. Additionally, we design the Prototypical Cross-frame Contrastive (PCC) loss, which ensures consistency in object-level features across frames, thereby significantly enhancing matching accuracy. CAVIS demonstrates superior performance over state-of-the-art methods on all benchmark datasets in video instance segmentation (VIS) and video panoptic segmentation (VPS). Notably, our method excels on the OVIS dataset, known for its particularly challenging videos. Project page: https://seung-hun-lee.github.io/projects/CAVIS/

Sungho Moon, Seunghun Lee, Jiwan Seo et al.

We propose Context-aware Video-text Alignment (CVA), a novel framework to address a significant challenge in video temporal grounding: achieving temporally sensitive video-text alignment that remains robust to irrelevant background context. Our framework is built on three key components. First, we propose Query-aware Context Diversification (QCD), a new data augmentation strategy that ensures only semantically unrelated content is mixed in. It builds a video-text similarity-based pool of replacement clips to simulate diverse contexts while preventing the ``false negative" caused by query-agnostic mixing. Second, we introduce the Context-invariant Boundary Discrimination (CBD) loss, a contrastive loss that enforces semantic consistency at challenging temporal boundaries, making their representations robust to contextual shifts and hard negatives. Third, we introduce the Context-enhanced Transformer Encoder (CTE), a hierarchical architecture that combines windowed self-attention and bidirectional cross-attention with learnable queries to capture multi-scale temporal context. Through the synergy of these data-centric and architectural enhancements, CVA achieves state-of-the-art performance on major VTG benchmarks, including QVHighlights and Charades-STA. Notably, our method achieves a significant improvement of approximately 5 points in Recall@1 (R1) scores over state-of-the-art methods, highlighting its effectiveness in mitigating false negatives.

Seunghun Lee, Jiwan Seo, Jeonghoon Kim et al.

Referring Video Object Segmentation (RVOS) aims to segment and track objects in videos based on natural language expressions, requiring precise alignment between visual content and textual queries. However, existing methods often suffer from semantic misalignment, largely due to indiscriminate frame sampling and supervision of all visible objects during training -- regardless of their actual relevance to the expression. We identify the core problem as the absence of an explicit temporal learning signal in conventional training paradigms. To address this, we introduce MeViS-M, a dataset built upon the challenging MeViS benchmark, where we manually annotate temporal spans when each object is referred to by the expression. These annotations provide a direct, semantically grounded supervision signal that was previously missing. To leverage this signal, we propose Temporally Grounded Learning (TGL), a novel learning framework that directly incorporates temporal grounding into the training process. Within this frame- work, we introduce two key strategies. First, Moment-guided Dual-path Propagation (MDP) improves both grounding and tracking by decoupling language-guided segmentation for relevant moments from language-agnostic propagation for others. Second, Object-level Selective Supervision (OSS) supervises only the objects temporally aligned with the expression in each training clip, thereby reducing semantic noise and reinforcing language-conditioned learning. Extensive experiments demonstrate that our TGL framework effectively leverages temporal signal to establish a new state-of-the-art on the challenging MeViS benchmark. We will make our code and the MeViS-M dataset publicly available.

Woo Kyoung Han, Yongjun Lee, Byeonghun Lee et al.

Despite significant advances in learning-based lossy compression algorithms, standardizing codecs remains a critical challenge. In this paper, we present the JPEG Processing Neural Operator (JPNeO), a next-generation JPEG algorithm that maintains full backward compatibility with the current JPEG format. Our JPNeO improves chroma component preservation and enhances reconstruction fidelity compared to existing artifact removal methods by incorporating neural operators in both the encoding and decoding stages. JPNeO achieves practical benefits in terms of reduced memory usage and parameter count. We further validate our hypothesis about the existence of a space with high mutual information through empirical evidence. In summary, the JPNeO functions as a high-performance out-of-the-box image compression pipeline without changing source coding's protocol. Our source code is available at https://github.com/WooKyoungHan/JPNeO.

Woo Kyoung Han, Byeonghun Lee, Hyunmin Cho et al.

We quantify the upper bound on the size of the implicit neural representation (INR) model from a digital perspective. The upper bound of the model size increases exponentially as the required bit-precision increases. To this end, we present a bit-plane decomposition method that makes INR predict bit-planes, producing the same effect as reducing the upper bound of the model size. We validate our hypothesis that reducing the upper bound leads to faster convergence with constant model size. Our method achieves lossless representation in 2D image and audio fitting, even for high bit-depth signals, such as 16-bit, which was previously unachievable. We pioneered the presence of bit bias, which INR prioritizes as the most significant bit (MSB). We expand the application of the INR task to bit depth expansion, lossless image compression, and extreme network quantization. Our source code is available at https://github.com/WooKyoungHan/LosslessINR

Minsu Kim, Jaewon Lee, Byeonghun Lee et al.

Existing frameworks for image stitching often provide visually reasonable stitchings. However, they suffer from blurry artifacts and disparities in illumination, depth level, etc. Although the recent learning-based stitchings relax such disparities, the required methods impose sacrifice of image qualities failing to capture high-frequency details for stitched images. To address the problem, we propose a novel approach, implicit Neural Image Stitching (NIS) that extends arbitrary-scale super-resolution. Our method estimates Fourier coefficients of images for quality-enhancing warps. Then, the suggested model blends color mismatches and misalignment in the latent space and decodes the features into RGB values of stitched images. Our experiments show that our approach achieves improvement in resolving the low-definition imaging of the previous deep image stitching with favorable accelerated image-enhancing methods. Our source code is available at https://github.com/minshu-kim/NIS.

Dahoon Park, Kon-Woo Kwon, Sunghoon Im et al.

In this paper, we present Zero-data Based Repeated bit flip Attack (ZeBRA) that precisely destroys deep neural networks (DNNs) by synthesizing its own attack datasets. Many prior works on adversarial weight attack require not only the weight parameters, but also the training or test dataset in searching vulnerable bits to be attacked. We propose to synthesize the attack dataset, named distilled target data, by utilizing the statistics of batch normalization layers in the victim DNN model. Equipped with the distilled target data, our ZeBRA algorithm can search vulnerable bits in the model without accessing training or test dataset. Thus, our approach makes the adversarial weight attack more fatal to the security of DNNs. Our experimental results show that 2.0x (CIFAR-10) and 1.6x (ImageNet) less number of bit flips are required on average to destroy DNNs compared to the previous attack method. Our code is available at https://github. com/pdh930105/ZeBRA.

Seokju Lee, Sunghoon Im, Stephen Lin et al.

We present an end-to-end joint training framework that explicitly models 6-DoF motion of multiple dynamic objects, ego-motion and depth in a monocular camera setup without supervision. Our technical contributions are three-fold. First, we highlight the fundamental difference between inverse and forward projection while modeling the individual motion of each rigid object, and propose a geometrically correct projection pipeline using a neural forward projection module. Second, we design a unified instance-aware photometric and geometric consistency loss that holistically imposes self-supervisory signals for every background and object region. Lastly, we introduce a general-purpose auto-annotation scheme using any off-the-shelf instance segmentation and optical flow models to produce video instance segmentation maps that will be utilized as input to our training pipeline. These proposed elements are validated in a detailed ablation study. Through extensive experiments conducted on the KITTI and Cityscapes dataset, our framework is shown to outperform the state-of-the-art depth and motion estimation methods. Our code, dataset, and models are available at https://github.com/SeokjuLee/Insta-DM .

Seokju Lee, Sunghoon Im, Stephen Lin et al.

We present an end-to-end joint training framework that explicitly models 6-DoF motion of multiple dynamic objects, ego-motion and depth in a monocular camera setup without supervision. Our technical contributions are three-fold. First, we propose a differentiable forward rigid projection module that plays a key role in our instance-wise depth and motion learning. Second, we design an instance-wise photometric and geometric consistency loss that effectively decomposes background and moving object regions. Lastly, we introduce a new auto-annotation scheme to produce video instance segmentation maps that will be utilized as input to our training pipeline. These proposed elements are validated in a detailed ablation study. Through extensive experiments conducted on the KITTI dataset, our framework is shown to outperform the state-of-the-art depth and motion estimation methods. Our code and dataset will be available at https://github.com/SeokjuLee/Insta-DM.

Wonhyeok Choi, Shutong Ding, Minwoo Choi et al.

Diffusion policies have emerged as a powerful approach for robotic control, demonstrating superior expressiveness in modeling multimodal action distributions compared to conventional policy networks. However, their integration with online reinforcement learning remains challenging due to fundamental incompatibilities between diffusion model training objectives and standard RL policy improvement mechanisms. This paper presents the first comprehensive review and empirical analysis of current Online Diffusion Policy Reinforcement Learning (Online DPRL) algorithms for scalable robotic control systems. We propose a novel taxonomy that categorizes existing approaches into four distinct families--Action-Gradient, Q-Weighting, Proximity-Based, and Backpropagation Through Time (BPTT) methods--based on their policy improvement mechanisms. Through extensive experiments on a unified NVIDIA Isaac Lab benchmark encompassing 12 diverse robotic tasks, we systematically evaluate representative algorithms across five critical dimensions: task diversity, parallelization capability, diffusion step scalability, cross-embodiment generalization, and environmental robustness. Our analysis identifies key findings regarding the fundamental trade-offs inherent in each algorithmic family, particularly concerning sample efficiency and scalability. Furthermore, we reveal critical computational and algorithmic bottlenecks that currently limit the practical deployment of online DPRL. Based on these findings, we provide concrete guidelines for algorithm selection tailored to specific operational constraints and outline promising future research directions to advance the field toward more general and scalable robotic learning systems.

Kyumin Hwang, Wonhyeok Choi, Kiljoon Han et al.

Recent foundation models demonstrate strong generalization capabilities in monocular depth estimation. However, directly applying these models to Full Surround Monocular Depth Estimation (FSMDE) presents two major challenges: (1) high computational cost, which limits real-time performance, and (2) difficulty in estimating metric-scale depth, as these models are typically trained to predict only relative depth. To address these limitations, we propose a novel knowledge distillation strategy that transfers robust depth knowledge from a foundation model to a lightweight FSMDE network. Our approach leverages a hybrid regression framework combining the knowledge distillation scheme--traditionally used in classification--with a depth binning module to enhance scale consistency. Specifically, we introduce a cross-interaction knowledge distillation scheme that distills the scale-invariant depth bin probabilities of a foundation model into the student network while guiding it to infer metric-scale depth bin centers from ground-truth depth. Furthermore, we propose view-relational knowledge distillation, which encodes structural relationships among adjacent camera views and transfers them to enhance cross-view depth consistency. Experiments on DDAD and nuScenes demonstrate the effectiveness of our method compared to conventional supervised methods and existing knowledge distillation approaches. Moreover, our method achieves a favorable trade-off between performance and efficiency, meeting real-time requirements.

Jihun Park, Jongmin Gim, Kyoungmin Lee et al.

We present Text-driven object-centric style editing model named Style-Editor, a novel method that guides style editing at an object-centric level using textual inputs. The core of Style-Editor is our Patch-wise Co-Directional (PCD) loss, meticulously designed for precise object-centric editing that are closely aligned with the input text. This loss combines a patch directional loss for text-guided style direction and a patch distribution consistency loss for even CLIP embedding distribution across object regions. It ensures a seamless and harmonious style editing across object regions. Key to our method are the Text-Matched Patch Selection (TMPS) and Pre-fixed Region Selection (PRS) modules for identifying object locations via text, eliminating the need for segmentation masks. Lastly, we introduce an Adaptive Background Preservation (ABP) loss to maintain the original style and structural essence of the image's background. This loss is applied to dynamically identified background areas. Extensive experiments underline the effectiveness of our approach in creating visually coherent and textually aligned style editing.

Wonhyeok Choi, Mingyu Shin, Sunghoon Im

Monocular 3D object detection poses a significant challenge due to the lack of depth information in RGB images. Many existing methods strive to enhance the object depth estimation performance by allocating additional parameters for object depth estimation, utilizing extra modules or data. In contrast, we introduce a novel metric learning scheme that encourages the model to extract depth-discriminative features regardless of the visual attributes without increasing inference time and model size. Our method employs the distance-preserving function to organize the feature space manifold in relation to ground-truth object depth. The proposed (K, B, eps)-quasi-isometric loss leverages predetermined pairwise distance restriction as guidance for adjusting the distance among object descriptors without disrupting the non-linearity of the natural feature manifold. Moreover, we introduce an auxiliary head for object-wise depth estimation, which enhances depth quality while maintaining the inference time. The broad applicability of our method is demonstrated through experiments that show improvements in overall performance when integrated into various baselines. The results show that our method consistently improves the performance of various baselines by 23.51% and 5.78% on average across KITTI and Waymo, respectively.

Wonhyeok Choi, Mingyu Shin, Hyukzae Lee et al.

Real-time processing is crucial in autonomous driving systems due to the imperative of instantaneous decision-making and rapid response. In real-world scenarios, autonomous vehicles are continuously tasked with interpreting their surroundings, analyzing intricate sensor data, and making decisions within split seconds to ensure safety through numerous computer vision tasks. In this paper, we present a new real-time multi-task network adept at three vital autonomous driving tasks: monocular 3D object detection, semantic segmentation, and dense depth estimation. To counter the challenge of negative transfer, which is the prevalent issue in multi-task learning, we introduce a task-adaptive attention generator. This generator is designed to automatically discern interrelations across the three tasks and arrange the task-sharing pattern, all while leveraging the efficiency of the hard-parameter sharing approach. To the best of our knowledge, the proposed model is pioneering in its capability to concurrently handle multiple tasks, notably 3D object detection, while maintaining real-time processing speeds. Our rigorously optimized network, when tested on the Cityscapes-3D datasets, consistently outperforms various baseline models. Moreover, an in-depth ablation study substantiates the efficacy of the methodologies integrated into our framework.

Jihun Park, Jongmin Gim, Kyoungmin Lee et al.

We present a training-free style-aligned image generation method that leverages a scale-wise autoregressive model. While large-scale text-to-image (T2I) models, particularly diffusion-based methods, have demonstrated impressive generation quality, they often suffer from style misalignment across generated image sets and slow inference speeds, limiting their practical usability. To address these issues, we propose three key components: initial feature replacement to ensure consistent background appearance, pivotal feature interpolation to align object placement, and dynamic style injection, which reinforces style consistency using a schedule function. Unlike previous methods requiring fine-tuning or additional training, our approach maintains fast inference while preserving individual content details. Extensive experiments show that our method achieves generation quality comparable to competing approaches, significantly improves style alignment, and delivers inference speeds over six times faster than the fastest model.

Jihun Park, Kyoungmin Lee, Jongmin Gim et al.

We present Infinite-Story, a training-free framework for consistent text-to-image (T2I) generation tailored for multi-prompt storytelling scenarios. Built upon a scale-wise autoregressive model, our method addresses two key challenges in consistent T2I generation: identity inconsistency and style inconsistency. To overcome these issues, we introduce three complementary techniques: Identity Prompt Replacement, which mitigates context bias in text encoders to align identity attributes across prompts; and a unified attention guidance mechanism comprising Adaptive Style Injection and Synchronized Guidance Adaptation, which jointly enforce global style and identity appearance consistency while preserving prompt fidelity. Unlike prior diffusion-based approaches that require fine-tuning or suffer from slow inference, Infinite-Story operates entirely at test time, delivering high identity and style consistency across diverse prompts. Extensive experiments demonstrate that our method achieves state-of-the-art generation performance, while offering over 6X faster inference (1.72 seconds per image) than the existing fastest consistent T2I models, highlighting its effectiveness and practicality for real-world visual storytelling.

Seunghun Lee, Jiwan Seo, Minwoo Choi et al.

In this paper, we present Latest Object Memory Management (LOMM) for temporally consistent video instance segmentation that significantly improves long-term instance tracking. At the core of our method is Latest Object Memory (LOM), which robustly tracks and continuously updates the latest states of objects by explicitly modeling their presence in each frame. This enables consistent tracking and accurate identity management across frames, enhancing both performance and reliability through the VIS process. Moreover, we introduce Decoupled Object Association (DOA), a strategy that separately handles newly appearing and already existing objects. By leveraging our memory system, DOA accurately assigns object indices, improving matching accuracy and ensuring stable identity consistency, even in dynamic scenes where objects frequently appear and disappear. Extensive experiments and ablation studies demonstrate the superiority of our method over traditional approaches, setting a new benchmark in VIS. Notably, our LOMM achieves state-of-the-art AP score of 54.0 on YouTube-VIS 2022, a dataset known for its challenging long videos. Project page: https://seung-hun-lee.github.io/projects/LOMM/

Kyoungmin Lee, Jihun Park, Jongmin Gim et al.

We present a training-free framework for style-personalized image generation that controls content and style information during inference using a scale-wise autoregressive model. Our method employs a three-path design--content, style, and generation--each guided by a corresponding text prompt, enabling flexible and efficient control over image semantics without any additional training. A central contribution of this work is a step-wise and attention-wise intervention analysis. Through systematic prompt and feature injection, we find that early-to-middle generation steps play a pivotal role in shaping both content and style, and that query features predominantly encode content-specific information. Guided by these insights, we introduce two targeted mechanisms: Key Stage Attention Sharing, which aligns content and style during the semantically critical steps, and Adaptive Query Sharing, which reinforces content semantics in later steps through similarity-aware query blending. Extensive experiments demonstrate that our method achieves competitive style fidelity and prompt fidelity compared to fine-tuned baselines, while offering faster inference and greater deployment flexibility.

Sanggyun Ma, Wonjoon Choi, Jihun Park et al.

We present Bridging Geometric and Semantic (BriGeS), an effective method that fuses geometric and semantic information within foundation models to enhance Monocular Depth Estimation (MDE). Central to BriGeS is the Bridging Gate, which integrates the complementary strengths of depth and segmentation foundation models. This integration is further refined by our Attention Temperature Scaling technique. It finely adjusts the focus of the attention mechanisms to prevent over-concentration on specific features, thus ensuring balanced performance across diverse inputs. BriGeS capitalizes on pre-trained foundation models and adopts a strategy that focuses on training only the Bridging Gate. This method significantly reduces resource demands and training time while maintaining the model's ability to generalize effectively. Extensive experiments across multiple challenging datasets demonstrate that BriGeS outperforms state-of-the-art methods in MDE for complex scenes, effectively handling intricate structures and overlapping objects.

Wonhyeok Choi, Kyumin Hwang, Minwoo Choi et al.

Self-supervised monocular depth estimation (SSMDE) has gained attention in the field of deep learning as it estimates depth without requiring ground truth depth maps. This approach typically uses a photometric consistency loss between a synthesized image, generated from the estimated depth, and the original image, thereby reducing the need for extensive dataset acquisition. However, the conventional photometric consistency loss relies on the Lambertian assumption, which often leads to significant errors when dealing with reflective surfaces that deviate from this model. To address this limitation, we propose a novel framework that incorporates intrinsic image decomposition into SSMDE. Our method synergistically trains for both monocular depth estimation and intrinsic image decomposition. The accurate depth estimation facilitates multi-image consistency for intrinsic image decomposition by aligning different view coordinate systems, while the decomposition process identifies reflective areas and excludes corrupted gradients from the depth training process. Furthermore, our framework introduces a pseudo-depth generation and knowledge distillation technique to further enhance the performance of the student model across both reflective and non-reflective surfaces. Comprehensive evaluations on multiple datasets show that our approach significantly outperforms existing SSMDE baselines in depth prediction, especially on reflective surfaces.

Wonhyeok Choi, Kyumin Hwang, Wei Peng et al.

Self-supervised monocular depth estimation (SSMDE) aims to predict the dense depth map of a monocular image, by learning depth from RGB image sequences, eliminating the need for ground-truth depth labels. Although this approach simplifies data acquisition compared to supervised methods, it struggles with reflective surfaces, as they violate the assumptions of Lambertian reflectance, leading to inaccurate training on such surfaces. To tackle this problem, we propose a novel training strategy for an SSMDE by leveraging triplet mining to pinpoint reflective regions at the pixel level, guided by the camera geometry between different viewpoints. The proposed reflection-aware triplet mining loss specifically penalizes the inappropriate photometric error minimization on the localized reflective regions while preserving depth accuracy in non-reflective areas. We also incorporate a reflection-aware knowledge distillation method that enables a student model to selectively learn the pixel-level knowledge from reflective and non-reflective regions. This results in robust depth estimation across areas. Evaluation results on multiple datasets demonstrate that our method effectively enhances depth quality on reflective surfaces and outperforms state-of-the-art SSMDE baselines.

Minjun Kang, Jaesung Choe, Hyowon Ha et al.

Many mobile manufacturers recently have adopted Dual-Pixel (DP) sensors in their flagship models for faster auto-focus and aesthetic image captures. Despite their advantages, research on their usage for 3D facial understanding has been limited due to the lack of datasets and algorithmic designs that exploit parallax in DP images. This is because the baseline of sub-aperture images is extremely narrow and parallax exists in the defocus blur region. In this paper, we introduce a DP-oriented Depth/Normal network that reconstructs the 3D facial geometry. For this purpose, we collect a DP facial data with more than 135K images for 101 persons captured with our multi-camera structured light systems. It contains the corresponding ground-truth 3D models including depth map and surface normal in metric scale. Our dataset allows the proposed matching network to be generalized for 3D facial depth/normal estimation. The proposed network consists of two novel modules: Adaptive Sampling Module and Adaptive Normal Module, which are specialized in handling the defocus blur in DP images. Finally, the proposed method achieves state-of-the-art performances over recent DP-based depth/normal estimation methods. We also demonstrate the applicability of the estimated depth/normal to face spoofing and relighting.

Jaesung Choe, Sunghoon Im, Francois Rameau et al.

To reconstruct a 3D scene from a set of calibrated views, traditional multi-view stereo techniques rely on two distinct stages: local depth maps computation and global depth maps fusion. Recent studies concentrate on deep neural architectures for depth estimation by using conventional depth fusion method or direct 3D reconstruction network by regressing Truncated Signed Distance Function (TSDF). In this paper, we advocate that replicating the traditional two stages framework with deep neural networks improves both the interpretability and the accuracy of the results. As mentioned, our network operates in two steps: 1) the local computation of the local depth maps with a deep MVS technique, and, 2) the depth maps and images' features fusion to build a single TSDF volume. In order to improve the matching performance between images acquired from very different viewpoints (e.g., large-baseline and rotations), we introduce a rotation-invariant 3D convolution kernel called PosedConv. The effectiveness of the proposed architecture is underlined via a large series of experiments conducted on the ScanNet dataset where our approach compares favorably against both traditional and deep learning techniques.

Seunghun Lee, Sunghyun Cho, Sunghoon Im

In this paper, we present DRANet, a network architecture that disentangles image representations and transfers the visual attributes in a latent space for unsupervised cross-domain adaptation. Unlike the existing domain adaptation methods that learn associated features sharing a domain, DRANet preserves the distinctiveness of each domain's characteristics. Our model encodes individual representations of content (scene structure) and style (artistic appearance) from both source and target images. Then, it adapts the domain by incorporating the transferred style factor into the content factor along with learnable weights specified for each domain. This learning framework allows bi-/multi-directional domain adaptation with a single encoder-decoder network and aligns their domain shift. Additionally, we propose a content-adaptive domain transfer module that helps retain scene structure while transferring style. Extensive experiments show our model successfully separates content-style factors and synthesizes visually pleasing domain-transferred images. The proposed method demonstrates state-of-the-art performance on standard digit classification tasks as well as semantic segmentation tasks.

Seokju Lee, Sunghoon Im, Stephen Lin et al.

We present a method for decomposing the 3D scene flow observed from a moving stereo rig into stationary scene elements and dynamic object motion. Our unsupervised learning framework jointly reasons about the camera motion, optical flow, and 3D motion of moving objects. Three cooperating networks predict stereo matching, camera motion, and residual flow, which represents the flow component due to object motion and not from camera motion. Based on rigid projective geometry, the estimated stereo depth is used to guide the camera motion estimation, and the depth and camera motion are used to guide the residual flow estimation. We also explicitly estimate the 3D scene flow of dynamic objects based on the residual flow and scene depth. Experiments on the KITTI dataset demonstrate the effectiveness of our approach and show that our method outperforms other state-of-the-art algorithms on the optical flow and visual odometry tasks.

Sunghoon Im, Hae-Gon Jeon, Stephen Lin et al.

Multiview stereo aims to reconstruct scene depth from images acquired by a camera under arbitrary motion. Recent methods address this problem through deep learning, which can utilize semantic cues to deal with challenges such as textureless and reflective regions. In this paper, we present a convolutional neural network called DPSNet (Deep Plane Sweep Network) whose design is inspired by best practices of traditional geometry-based approaches for dense depth reconstruction. Rather than directly estimating depth and/or optical flow correspondence from image pairs as done in many previous deep learning methods, DPSNet takes a plane sweep approach that involves building a cost volume from deep features using the plane sweep algorithm, regularizing the cost volume via a context-aware cost aggregation, and regressing the dense depth map from the cost volume. The cost volume is constructed using a differentiable warping process that allows for end-to-end training of the network. Through the effective incorporation of conventional multiview stereo concepts within a deep learning framework, DPSNet achieves state-of-the-art reconstruction results on a variety of challenging datasets.

Sunghoon Im, Hae-Gon Jeon, In So Kweon

As demand for advanced photographic applications on hand-held devices grows, these electronics require the capture of high quality depth. However, under low-light conditions, most devices still suffer from low imaging quality and inaccurate depth acquisition. To address the problem, we present a robust depth estimation method from a short burst shot with varied intensity (i.e., Auto Bracketing) or strong noise (i.e., High ISO). We introduce a geometric transformation between flow and depth tailored for burst images, enabling our learning-based multi-view stereo matching to be performed effectively. We then describe our depth estimation pipeline that incorporates the geometric transformation into our residual-flow network. It allows our framework to produce an accurate depth map even with a bracketed image sequence. We demonstrate that our method outperforms state-of-the-art methods for various datasets captured by a smartphone and a DSLR camera. Moreover, we show that the estimated depth is applicable for image quality enhancement and photographic editing.